array of Float in device

I meet a problem in mallocing an array of 320000 floats in CudaDevice , while mallocing an array of 3200000 ints or doubles will be fine.
Before this I tried to used thrust::device_vector which size is 10000000, but now I can’t.
Does somebody else know the reason for this ?
I am waiting for you help.
Addition:Cuda 6.0,compute_20,sm_20

I believe your post might have a typo… one of your array sizes is a degree of magnitude higher than the other – 320,000 vs 3,200,000… which is the one that failed?

It would seem like you might be running out of device memory on your GPU… read the corresponding documentation/poll for memory usage as appropriate to make sure that isn’t the case. Also make sure you check for errors for the individual CUDA calls as appropriate.